Recognition and Evaluation of Cables for Automated Cutting in Dismantling Process

Electronic waste (e-waste) continues to increase annually. Automation of the dismantling process is essential for improving recycling rates, but high-mix, low-volume e-wastes make automation particularly challenging, and this leads to reliance on manual operations. The demand for robotic systems that operate without specialized positioning jigs or dismantling lines is increasing. To promote sustainable resource circulation, our aim is to achieve Automation of e-waste dismantling. For this purpose, we propose a novel wire recognition framework and evaluation metrics for wire-cutting tasks. We utilize a pre-trained instance segmentation model fine-tuned with a custom cable image dataset. Because of deformable linear object, cables are harder to recognize than rigid objects and tend to score lower on Average Precision (AP) and Average Recall (AR). Since precise recognition near cutting points is critical for cable-cutting tasks, we propose new evaluation metrics tailored to such operations. Experiments on condenser units demonstrate robust recognition accuracy for cable-cutting, even with overlapping cables. This study validates the effectiveness of the proposed recognition method and evaluation metrics in automating the dismantling process. Future work will address the detection of cutting points and motion generation for cable-cutting operations.